Upload
others
View
6
Download
0
Embed Size (px)
Citation preview
ulvacsim.paper.doc 08/20/99 p. 1 of 23
Dynamic Simulation of a Multichamber CVD Cluster Tool
N. Gupta(a), Y. Xu, L. Henn-Lecordier, T. Gougousi, J. N. Kidder, Jr., and G. W. Rubloff
Institute for Systems Research and Department of Materials and Nuclear EngineeringUniversity of Maryland, College Park, MD 20742-3285
Abstract
Dynamic, physically based simulation has proved very effective in representing the time-
dependent behavior of equipment, process, sensor and control systems. We have developed a
system-level dynamic simulator for an 8” CVD cluster tool (ULVAC-ERA1000). The simulator
incorporates models for equipment, process and control systems, representing time-dependent
system level behavior through the imposition of equipment controller recipes for pump-down,
process and venting cycles. Multistage pumping systems (Roots-mechanical and turbo-
mechanical) are modeled using compression ratios, nominal speeds, and volumes to reflect
actual behavior, validated against experiments on the Ulvac tool. The process simulator is
derived from the work of Hsieh et al. The Ulvac simulator is applied to optimization of
deposition rate and WF6 utilization, analysis of experimental data for recipe improvement, and
development and debugging of equipment controller software and operation.
a) Current address: Intel Corporation, Hillsboro, OR
ulvacsim.paper.doc 08/20/99 p. 2 of 23
IntroductionThe National Technology Roadmap for Semiconductors1 asserts strongly that continuing
progress in integrated circuit technology is increasingly dependent on enhanced equipment
productivity, involving costs of capital equipment, maintenance, and consumables, along with
improved equipment reliability and short time-to-repair. The importance of these factors is
underscored by the observations that equipment costs dominate the cost of new multi-billion
dollar manufacturing plants, and that focus on these issues has been formalized and
emphasized through the use of the concepts of cost-of-ownership (COO) and overall equipment
effectiveness (OEE).
Stimulated by the successes of technology modeling at the process, device, and circuit
levels, significant attention has been devoted to equipment or reactor modeling through
approaches such as computational fluid dynamics 2,3,4. These have been coupled through
explicit process models to reactor performance and the prediction of the evolution of
microfeature profiles5,6,7,8,9,10. However, the dynamics of equipment behavior through a process
cycle has been treated in only isolated cases, most notably to meet the challenge of rapid
thermal processing 11,12.
In this paper, we have studied the behavior of a multi-chamber cluster tool at the system
level. System-level dynamic simulation provides a powerful technique for studying the time-
dependant behavior of process, equipment, sensors and control systems of a manufacturing
tool. This has been shown by Lu et al.13,14 through simulation of polysilicon RTCVD processes.
Their results demonstrate that system level simulation allows the evaluation of not only the
deposition-related parameters, but also a broad range of system properties like equipment
performance, gas and heat flow conditions and process cycle time. We have developed a
similar model for a tungsten CVD cluster tool, which uses reduced order physical and chemical
models to describe the gas transport, heat transfer and deposition kinetics in the reactor. The
simulator captures the behavior of a full-scale manufacturing tool with multiple chambers (two
reactors, a load lock and a central wafer handler), and a multi-pump vacuum system. The gas
flow model for such a system must integrate the functioning of turbo and mechanical pumps in
a methodology consistent with behavior of a sequential pump system. The individual elements
representing equipment, process and control systems are integrated by the implementation of
ulvacsim.paper.doc 08/20/99 p. 3 of 23
process recipes which model the behavior of the cluster tool controller. The simulator has been
validated with experimental results, and is being used to investigate the dynamics and control of
deposition in tungsten CVD processes.
BackgroundSystem level dynamic simulation has been previously used by Lu et al.8 to study the
dynamics and control of polysilicon RTCVD. The RTCVD equipment consists of the gas
handling system, a single wafer reactor and reactor pumping station and a two-stage
differentially pumped QMS system. The simulator is composed of five elements. Since the
present work builds upon this structure, it is useful to understand the purpose of these
components in the polysilicon RTCVD simulator. The Process Recipe defines the status of the
valves, mass flow controllers (MFC’s), power to the heating lamps and overall process
conditions as a function of time. The Equipment Simulator uses the status of the valves, MFC’s
etc to calculate the gas flow and heat flow conditions in the reactor. The mathematical models
of the gas transport and heat transfer are based on reduced order mass and energy balance
equations. The outputs of the Equipment Simulator are the reactor pressure and the wafer
temperature. The Sensors and Control Systems simulator reads in this pressure and
temperature and compares them to the process parameters preset in the deposition recipe. It
then regulates the reactor pressure by changing the position of the throttle valve and the wafer
temperature by varying the power to the heating lamp, and thus the nominal process conditions
are established. The deposition kinetics are modeled in the Process Simulator which calculates
growth rates and film thickness, which are the final output of the integrated model.
Ulvac ERA-1000 Equipment DescriptionThe ULVAC-ERA 1000 at the Laboratory for Advanced Materials Processing, University of
Maryland at College Park, is a CVD cluster tool designed for tungsten deposition. It consists of
a gas handling system, the CVD reactor (including a load-lock, central wafer handler, two
reaction chambers and heating lamps), reactor pumping system and equipment controls. Fig. 1
shows the schematic layout of the cluster tool.
The gas handling system consists of mass flow controllers, which regulate the flow of the
reactant gases (WF6 and H2 for our deposition process) to the reaction chambers. The reaction
chambers are cylindrical in shape and are located symmetrically on either side of the central
wafer handler (CWH), which is then linked to a load-lock (LL). The load lock isolates the CWH
ulvacsim.paper.doc 08/20/99 p. 4 of 23
from the room ambient via a gate valve (V3). The CWH contains a robot arm, which takes the
wafer from the LL and deposits it on a wafer lifter pin inside the reaction chamber via a gate
valve (V1 or V2). The pressure in the reactor chamber is measured using a capacitance
manometer. The pressure during deposition is controlled using a feedback loop to continuously
adjust the throttle valve position on the reactor pumping station while maintaining a constant
gas flow to the reactor. During deposition, the wafer chuck rotates at a speed of 5 rpm to
ensure uniform heating.
The ULVAC cluster tool has a multistage pumping system. Each reactor is connected to
two sets of pumps, which perform different function during the process cycle. One is used for
chamber evacuation and consists of a turbo pump (TMP1 & 2) backed by a rotary vane pump
(RP4). The second set of pumps, each consisting of a Roots blower (RBP1 & 2) backed by a
rotary pump (RP1 & 2), is used to maintain constant pressure during the deposition process.
Reactor pumping is switched from one pump stack to the other by the use of gate valves. The
LL and the CWH are jointly connected to a Roots blower (RBP3) backed by a rotary vane pump
(RP3).
Overall Simulator StructureThe CVD cluster tool is modeled in three sections: the gas handling system, the reactor
(including LL, CWH and two process chambers), and the pumping stations. Each section is
modeled in terms of three processes: gas flow, heat transfer and the reaction chemistry. Gas
flow calculations for the ULVAC cluster tool are based on mass balance in each piece of the
equipment. We have assumed ideal mixing between gases in the reactor to simplify our
computations. The total number of molecules in each chamber must equal the molecules input
to the chamber minus the molecules pumped out. For the reaction chambers the mathematical
expression is
where Pinit is the initial (base) pressure of the reaction chamber in Torr, Pfin is the final pressure,
Qin is the total throughput of gas (in Torr-liter/sec) entering the reactor, Qout is the throughput of
( ) )1(1
__ dtQQQQV
PP depletionreactantproductsreactionoutininitfin ∫ −+−+=
ulvacsim.paper.doc 08/20/99 p. 5 of 23
gas leaving the reactor, Qreaction_products is the throughput introduced due to product generation
and Qreatant_depletion denotes the gas throughput removed from the chamber due to reactant
depletion. The total and partial pressures of the reactor gases can thus be calculated at every
iteration of the simulation.
Equipment gas flow modeling
Previous models developed for system-level simulation of CVD reactors have assumed a
uniform pumping speed for the reactor pumping station, which is a good approximation of the
vacuum system behavior when the reactor is evacuated by a single pump. However, the
complexity of the pumping system on the ULVAC cluster tool necessitates a more detailed
modeling of the mechanical and turbo pumps. Inclusion of factors like the pump start-up time or
the gradual decay of pumping efficiency with decreasing pressure allows a better
representation of the equipment state.
Heat transfer modeling
Heat flow calculations in our simulation are based on reduced-order energy balance
equations. The wafer is assumed to absorb power from the heating lamps primarily by radiation,
with corrections for conductive and convective losses. Wafer temperature is regulated by
controlling the power to the heating lamps. The energy balance for the wafer can be expressed
as:
Where aQA corresponds to the power absorbed by the wafer, 2eAσT4w is the power radiated by
the wafer, Wcc is the power lost by conduction and convection and mCp (dTw/dt) is the power
consumed for heating the wafer; a is the wafer absorptivity; Q is the incoming lamp radiation; e
is the wafer emissivity; A is the wafer area, σ is the Stefan-Boltzmann constant; Tw is the wafer
temperature; m is the mass of the wafer; Cp is the specific heat of the wafer material and dTw/dt
is the rate of heating of the wafer. Wcc is modeled as a linear function of the difference between
the wafer temperature and the ambient temperature.
)2(2 4
dt
dTmCWTeAaQA w
pccw =−− σ
ulvacsim.paper.doc 08/20/99 p. 6 of 23
Chemical kinetics modeling
The chemical kinetics for the film deposition are calculated by the Process Simulator,
using as inputs the partial pressures of the process gases and the wafer temperature. Due to
the cold wall configuration of the reactor, and the relatively low temperatures of operation (400-
550 C), we have excluded gas phase reactions in the kinetic model. The surface reactions for
tungsten deposition from WF6 and H2 gas have been adapted from Hsieh’s model 9,10 for the
reduction of WF6 by SiH4 followed by H2.
Process recipe modeling
Equipment operation and process dynamics are integrated by the implementation of a
process recipe, which models the process sequence and the timing of the equipment controller.
This helps to understand the complex dynamics of a manufacturing tool and to explain
irregularities in equipment behavior by allowing one to study a broad range of system
properties.
Simulator developmentGas flow in a multistage pumping system
Mechanical pumps like the rotary vane pump (RP1, 2, 3 & 4) operate in the low to medium
vacuum region and reach full speed almost instantly on being turned on at atmospheric
pressure. Fore pumps like Roots blowers (RBP1, 2 & 3) or turbo pumps (TMP1 & 2) have a
longer start-up time because they do not reach their optimal speed unless the backing pressure
is less than 10 mtorr. Thus they are typically backed by a mechanical pump (rotary vane pump
in this case) to pump down the fore line to the mtorr range 15. We have taken into account this
time delay by empirically modeling the ramp-up of pump speed with decreasing pressure in the
foreline. The variation of pumping speed with pressure at the high end of the fore pump
operation range is shown in Fig. 2
At the low end of the pump operation range, all mechanical and turbo pumps (which use
variations of rotating fans or turbine blades to pump gas molecules from one end of the pump
to another) reach a limit in their pumping capacity before they reach their lowest pressure. The
lowest pressure ideally achievable by such pumps is a called the base pressure (usually
specified by the pump manufacturer) and is the parameter taken into consideration when
designing vacuum systems. However, when the inlet of the pump reaches a low pressure, gas
ulvacsim.paper.doc 08/20/99 p. 7 of 23
particles start backstreaming from the pump exhaust towards the inlet. Once the flux of the
backstreaming molecules equals the flux in the forward direction, a steady state pressure,
which is the actual base pressure is established.
This backstreaming of gas particles is measured in terms of a compression ratio, which is
defined as the ratio of the outlet pressure to the inlet pressure of a pump 11. The compression
ratio depends on the speed and design of the turbine blades, and varies with the size of the
species being pumped, i.e. light gases like hydrogen and helium have much smaller
compression ratios than heavy gases like argon, and are harder to pump on. If the compression
ratio is constant for a pump for a given process, lower base pressures can be achieved by
lowering the exhaust pressure, usually done by using a backing pump. In processes using a
combination of gases, the ultimate pressure is determined by the compression ratio of the
lighter gases.
The pressure-dependent pumping speed for each pump in the lower pressure regime has
been modeled in terms of the compression ratio of the pump. The effective gas throughput for a
pump is a combination of flow in the “forward” or inlet-to-exhaust direction and the flow due to
backstreaming particles.
Throughput in the forward direction: Qf (Torr-liter/sec) = (Ppump inlet – Pbase) * PS (3)
Throughput due to backstream: Qb (Torr-liter/sec) = (Poutlet– Ppump inlet) * PS / CR (4)
(in the reverse direction)
Effective throughput: Qeff = [(Ppump inlet – Pbase) - (Poutlet– Ppump inlet) * 1/ CR]*PS (5)
where Ppump inlet is the pressure at the chamber being evacuated. For the fore pumps (RBP and
TMP), Ppump inlet is the pressure at the reaction chamber, and for the rotary vane backing
mechanical pump (RP), it is the pressure at the outlet of it’s fore pump. Poutlet is the pressure at
the exhaust of the pump. The fore pumps exhaust into the inlet of the backing pumps, which in
turn pump out to atmosphere. Pbase is the theoretical base pressure of the pump, which would
have been in the absence of backstreaming gas. PS is the maximum speed of the pump, which
it achieves at its optimal pressure range. CR is the compression ratio, which is usually specified
by the manufacturer for a particular gas.
ulvacsim.paper.doc 08/20/99 p. 8 of 23
Thus the effective pumping speed in the low pressure end of the pump’s operating range
is not PS but is given by
Modeling the transient behavior of process pumps has a significant impact on the total process
cycle time in a multi-pump setup. Because of the sequential operation of the pumps, the start-
up time for the fore pumps depends on the lowest pressure achieved by the mechanical
backing pumps. Secondly, the conductance of the process pumps is limited by the fact that
during tungsten deposition, the bulk of the reactor gas is composed of hydrogen, which has a
very small compression ratio. This could affect the time response of the throttle valve to
pressure changes during the process.
Chemical Kinetics Model for W CVD
We consider here the Hsieh model 9,10 for the surface reduction of WF6 by (SiH4 + H2),
which has three competing reaction pathways of the Eley-Rideal type. The first pathway
describe the SiH4 reduction of WF6, the second pathway the Si reduction of WF6, and the third
the H2 reduction of WF6
We have modified the Eley-Rideal mechanism to obtain a three step kinetic model for the
surface reduction of WF6 by H2. In the first step, WF6 gas is reduced by surface silicon to form an
initial seed layer of tungsten. In the next step, WF6 gas adsorbs on to active tungsten sites.
Finally, adsorbed WF6 is reduced by H2 to deposit W metal. The relevant equations are given
below:
Nucleation: WF6 + 3/2 Si* È W* + 3/2 SiF4 (g) R1 = 0.3 JWF6 θSi (7)
Adsorption: W* + WF6 È WF6* R2 = 0.48 JWF6 θW (8)
Reduction: WF6* + 3H2 È W(s) + 6HF R3 = kH2 [PH2]0.5 θWF6 (9)
Site conservation: θSi + θWF6 + θW*=1 (10)
where kH2 (mol/cm2/s.Torr0.5) = 0.0083 exp[-8800/T(K)] is an equipment dependent empirical
constant depending on the process temperature. JWF6 is the molar flux of WF6 gas, [PH2] is the
)6(]1
)(
)(1[)/( PS
CRPP
PPslitersPS
baseinletpump
inletpumpoutleteff ××
−−
−=
ulvacsim.paper.doc 08/20/99 p. 9 of 23
partial pressure of H2 in the reactor and θSi, θW and θWF6 are the fractions of the total surface
active sites covered by Si*, W* and WF6* respectively. The rates of the first two reactions are
proportional to the sticking probabilities (0.3 and 0.48 respectively) of the impinging gas on the
active site.
The dynamic rates of the reactions are calculated in terms of the instantaneous
concentrations of the reactant gases and surface sites. In turn, the surface coverage of Si, W
and WF6 can be calculated by integrating over time the rates of all the reaction steps that
produce or consume the particular species, at each iteration step. The net deposition rate Rnet
equals (R1+R3), since both reactions deposit tungsten metal on the surface. The total film
thickness is obtained by integrating the growth rate over time.
Initially, the surface coverage (θSi) of the Si sites is 1. Due to low activation energy, the
nucleation reaction takes place at a very high rate and rapidly drives θSi to zero. For simplicity,
diffusion of Si atoms through the W film is not included in the simulator. Once the initial seed
layer of W* is deposited covering all available Si sites, R1 becomes zero. Shortly after, R2 and
R3 reach a steady-state equilibrium, i.e. R2=R3, and the bulk of the W film is deposited.
The deposition of W films using H2 reduction of WF6 has traditionally been regarded as a
slow and inefficient process. According to the simulator, in the pressure range of 0.1-0.5 Torr
and temperatures around 400-500 C, less than 10% of the WF6 is deposited on the wafer. It is
known that the deposition rate as well as the WF6 conversion rate can be significantly
enhanced with the introduction of small amounts of silane (SiH4).
Process Recipe for Model Integration
The various elements for gas and heat flow, deposition kinetics etc. have been integrated
into a coherent unit in order to achieve a system-level representation of the cluster tool. This
has been done in two steps. First, the controller command sequence and interlocks were
duplicated in as much detail as possible, to allow us to simulate the functioning of a full-scale
manufacturing tool. Thus the Process Recipe module defines valve states and interlocks and
MFC states, which represent equipment states, such as venting the load lock and wafer handler
to load the wafer, pumping down all chambers prior to deposition, opening up mass flow
controllers and turning on the heating lamp to begin deposition. The status of valves, MFC’s
and heating lamps is used to compute gas and heat transport in the system.
ulvacsim.paper.doc 08/20/99 p. 10 of 23
Secondly, to simulate a deposition process, a number of process parameters must be
entered by the user. Numerical values for the flow rates of the process gases, process
pressure, temperature and timing for each step of the deposition recipe (if it is a multi-step
recipe) form the complete set of inputs. Once the simulation is initiated, the wafer is loaded into
the reactor via the load lock and the central wafer handler. When the reactor is ready for
deposition, the gas flow, pressure and temperature control systems implement a series of
actions to achieve and maintain the set values of deposition parameters. A detailed modeling of
the equipment controller aids significantly in understanding and analyzing the dynamics of the
cluster tool at every step of its operation.
ValidationThe various modules of the ULVAC cluster tool simulator have been validated
independently to create a tool which can be directly used for experimental design and process
control strategies.
To validate the simulation results for equipment dynamics we developed a data
acquisition system (using DAQ cards from Computer Boards) to read relevant analog and
digital voltage signals (for process pressure, valve status and conductance, wafer temperature
etc) from the equipment controller. Once the voltage signals were scaled to range of the
respective parameter, they were directly compared to the simulation results. Fig. 3 presents
simulation results for the reactor pressure and the status (on/off) for valves V11 and V41 (see
Fig. 1) as a function of time. Data is shown for the wafer loading part of the process cycle. The
valve connecting the reactor to the process pumps (V11) is closed when the wafer is being
loaded in through the load lock and central wafer handler. Once the wafer is loaded the reactor
is pumped down using the turbo pumps (V41) to the 10-6 Torr range to remove all
contaminants. After the successful completion of this step, valve V11 opens up as deposition
commences. This detailed modeling of equipment behavior has been instrumental in analysis
and optimization of system behavior.
In order to validate the accuracy of implementation of Hsieh’s model in the simulator, the
simulated results for film growth rate have been compared to the experimental results reported
by Rosler and co-workers16. Fig. 4 shows our simulation results for the tungsten deposition
process, which seem to be in good agreement with experimentally measured growth rates.
ulvacsim.paper.doc 08/20/99 p. 11 of 23
Applications of Dynamic Cluster Tool SimulatorOptimization of Deposition Processes
An experimentally validated dynamic simulator proves to be an invaluable tool for process
recipe improvement and optimization. A primary application of the ULVAC cluster tool has been
in defining a process regime for tungsten deposition, which would optimize both the growth rate
and reactant consumption. To target an optimal process pressure and reactant composition
which would generate improved process recipes, we simulated deposition processes for
different H2:WF6 flow rate ratios and process pressures (0.2 and 0.5 torr), for the temperature
range 350-600 °C. In each case the flow rate of H2 is kept fixed at 200 sccm, and the results
are shown in Fig. 5.
Simulation results showed that although the conversion rates for WF6 are better for high
ratios of H2:WF6, deposition rates can be enhanced by moving towards a hydrogen limited
regime (low H2:WF6 ratio), which is also in favor of uniform W thin film. Increasing the process
pressure benefits both the film growth rate and the conversion ratio. This behavior can be
understood as arising from a longer residence time for the reactant gases in the reactor. For
the same reactants inlet rates, if the process pressure is increased, the throttle valve
conductance must decrease to reach the set-point pressure. This means that the gas
molecules remain in the reactor for a longer time, increasing the probability of their impinging
on the wafer surface and thus increase the growth rate and the conversion ratio as well. Once
simulation results have been used to define the desired process regime, smaller and better-
focussed sets of actual experiments may be designed for optimal process conditions.
Equipment Diagnostics and Fault Detection
A dynamic simulator can generate the time-dependent response of all the process and
equipment related parameters through the process cycle. This can be compared against the
signals from the equipment sensors to detect and analyze dynamic details of system operation.
We have used this capability explicitly to debug and better understand the subtleties of the
cluster tool, particularly the relative timing of actuation events involving valves, flow controllers,
etc. The integration of multiple equipment state signals, and their time-synchronous
combination with chemical sensor signals, has revealed important but subtle feature in the
equipment dynamics, while the dynamic tool simulator has been the primary vehicle which
enables interpretation and understanding of the system dynamics.
ulvacsim.paper.doc 08/20/99 p. 12 of 23
An example of this simulator application is illustrated in Fig. 6. The experimental data
showed that at the end of the deposition process (when the wafer was being transferred from
the reactor back to the central wafer handler through the slit valve) the pressure in the reactor
shot up briefly in a sharp spike. This seemed abnormal since there should have been no
pressure gradient between the reactor and the CWH during wafer transfer. In an attempt to
explain this phenomenon, we modified the Process Recipe module to simulate the following
sequence: during the removal of the wafer from the reactor, the mass flow controllers would
simultaneously open up completely for a second to flush out all the process gases. The
resulting brief pressure spike in the reactor corresponded closely to experimental data obtained
from the cluster tool, as seen in Fig. 6. Thus it is the combination of sensor integration and
dynamic simulation, which provides a powerful tool for system analysis and potential
improvement.
Test-bed for Equipment Controller Development
Finally, we have begun to exploit the dynamic simulator as a vehicle for equipment
controller development. In order to enhance the versatility of equipment control for
implementing various approaches to advanced process control, we have been bringing up a
new tool control system based on a Brooks Automation platform. Just as in the design of any
new equipment control system, a major challenge is to develop and validate the software
recipes in the control system. If the real equipment must be the test-bed for control software
debugging, significant cost and delay in equipment usage will result. The object here is to
exploit the dynamic simulator for validating and debugging control software, rather than
requiring dedication of real hardware to the task.
The ULVAC simulator has been designed so that the Process Recipe module can be
isolated and/or disconnected from the equipment and process modules, thereby allowing the
simulator to be operated in a fully manual way, i.e., one actuator (valve, flow controller, etc.) at
a time. This makes the simulator a "virtual tool" operable by an external agent (i.e., a user or
some other control system). We added a real-time interface (hardware interface cards and
their software support) to the VisSim simulator, so that simulator PC could sense and respond
to electrical actuation signals from the Brooks controller, and also return electrical signals from
the simulator to the controller. We were then able to use the Brooks controller, and its software
for process recipes, to control the ULVAC dynamic simulator, and thereby to see how the
system - represented faithfully by the dynamic simulator - behaved under the control system
process recipes in the Brooks controller.
ulvacsim.paper.doc 08/20/99 p. 13 of 23
Summary and ConclusionsWe have developed a system level simulator for the deposition of tungsten in a multi-
chamber CVD cluster tool. The overall architecture provides for the incorporation of
manufacturing equipment dynamics as well as reduced order process models in a coherent
manner. The individual modules have been integrated through the imposition of equipment
controller recipes for the cluster tool operation. The Process Recipe determines the state of the
valves, flow controllers, and heating lamps at each step of the process cycle, and defines a
sequence of actions for the equipment to achieve them. In this way, the dynamics of a
manufacturing process tool have been simulated at the system level. Validation of the dynamic
simulator has been accomplished by comparison with experimental data for individual simulator
components and more complex system-level behavior.
This dynamic simulator extends previous work 13,14 by providing a more sophisticated
representation of pumping systems, specifically multistage pumping arrangements, and by
representing a full multichamber cluster tool. In this work, we have seen its applications to initial
process optimization, interpreting important subtleties of equipment dynamics, and testing of
control system software and operation. Given the manufacturing importance attributed to the
details of equipment behavior - and particularly dynamics - this research direction appears of
substantial promise to impact a variety of equipment-centered manufacturing issues.
AcknowledgementsWe appreciate valuable discussions with B. Conaghan, S. Butler, E. Hall, and M.
Hanssmann. This work has been supported in part by Texas Instruments, the Semiconductor
Research Corporation, and the National Science Foundation.
ulvacsim.paper.doc 08/20/99 p. 14 of 23
References 1 National Technology Roadmap for Semiconductors 1997 (SIA Semiconductor Industry
Association, San Jose, CA, www.semichips.org, 1998).2 E. Sachs, G. H. Prueger, and R. Guerrieri, "An equipment model for LPCVD", IEEE Trans.
Semiconduct. Manuf. 5, p. 3 (Feb, 1992).3 C. R. Kleijn, "The modeling of LPCVD in single-wafer reactors as a tool for process
optimization and equipment design", J. de Physique IV, 1 (C2), pp. 19-31 (Sept. 1991).4 C. R. Kleijn, C. J. Hoogendoorn, A. Hasper, J. Holleman, and J. Middelhoek, "Transport
Phenomena in Tungsten LPCVD in a Single-Wafer Reactor", J. Electrochem. Soc. 138 (2)(Feb., 1991).
5 T. S. Cale, V. Mahadev, Z. Tang, G. Rajagopalan, L. J. Borucki, “Topography evolution duringsemiconductor processing”, Proceedings of NATO Advanced Study Institute, vol. 109-24,p. 613, (1997)
6 J. P. McVittie, J. C. Rey, A. J. Bariya, M. M. IslamRaja, L. Y. Cheng, S. Ravi, K. C. Saraswat,Proceedings SPIE Symposium on Advanced Techniques for Integrated Circuit Pocessing,vol. 1392, p. 126, (1990)
7 A. Hasper, J. Holleman, J. Middelhoek, C. R. Kleijn, and C. J. Hoogendoorn, "Modeling andOptimization of the Step Coverage of Tungsten LPCVD in Trenches and Contact Holes",J. Electrochem. Soc. 138 (6), 1728-38 (June 1991).
8 T. S. Cale, M. K. Jain, and G. B. Raupp, "Programmed Rate Processing to IncreaseThroughput in LPCVD", J. Electrochem. Soc. 137 (5), pp. 1526-33 (May 1990).
9 J. J. Hsieh, “Kinetic model for the chemical vapor deposition of tungsten in the silanereduction process”, J. Vac. Sci. Technol A 11 (6), pp. 3040-46, (Nov/Dec 1993).
10 J. J. Hsieh, "Influence of surface-activated reaction kinetics on low-pressure chemical vapordeposition conformality over micro features", J. Vac. Sci. Technol. A 11 (1), pp. 78-86(Jan/Feb 1993).
11 K. F. Jensen, and D. B. Graves, “Modeling and analysis of low pressure CVD reactors”, J.Electrochem. Soc. 130, p. 1950, (Sept. 1983)
12 F. Y. Sorrell, J. A. Harris, R. S. Gyurcsik, IEEE Trans. Semiconduct. Manuf., vol. 3, p. 183,(Nov. 1990)
13 G. Lu, M. Bora, L. L. Tedder, G. W. Rubloff, “Integrated dynamic simulation of rapid thermalchemical vapor deposition of polysilicon”, IEEE Trans. Semiconduct. Manuf. Vol. 11, p.63, (Feb. 1998)
14 G. Lu, M. Bora, G. W. Rubloff, “Polysilicon RTCVD process optimization for environmentally-conscious manufacturing”, vol. 10, p. 390, (Aug. 1997).
15 P. A. Redhead, J. P. Hobson, and E. V. Kornelson, “The Physical Basis of UltrahighVacuum”, Chapman and Hall, 1968.
ulvacsim.paper.doc 08/20/99 p. 15 of 23
16 R. S. Rosler, J. Mendonca, and M. J. Rice, “Tungsten chemical vapor depositioncharacteristics using SiH4 in a single wafer system”, J. Vac. Sci Technol. B 6 (6), p. 1721(1998).
ulvacsim.paper.doc 08/20/99 p. 16 of 23
Figure Captions
Figure 1. Schematic draws of the Ulvac ERA-1000 CVD cluster tool. Reactors 1 and 2 are
pumped by Roots pumps RBP1 and RBP2 backed by mechanical pumps RP1 and
RP2 during processing, while they are pumped to base pressures prior to process by
turbopumps TMP1 and TMP2, backed by mechanical pump RP4. Mechanical pump
RP3 is used to evacuate the load lock. The various chambers are isolated from each
other by slit (gate) valves V1, V2, and V3.
Figure 2. Variation of pumping speed with system pressure for turbo and mechanical pumps.
Figure 3. Comparison of reactor pressure and valve state signals during the initial part of the
process cycle, i.e., wafer loading into the reactor for deposition. Data points are from
dynamic simulation, while solid lines are experimental measurements using
equipment state signal integration under LabView. Voltage signals for valve states
(∼3.0V) are scaled up by 100X for clarity. Valve state data from the simulation is
scaled vertically to match amplitude of experimental signals.
Figure 4. Comparison of simulated deposition rate with experimental results reported by Rosler
et. al. 16.
Figure 5. Analysis of deposition rate and WF6 conversion vs. temperature for different reactant
ratios (H2:WF6 flow rate ratio = 20:1, 10:1, 5:1) for process pressures (a) 0.2 Torr
and (b) 0.5 Torr.
Figure 6. Dynamic simulation (triangles) and experimental data (smooth curve) for pressure
burst observed in reactor during post-deposition wafer unload.
ulvacsim.paper.doc 08/20/99 p. 17 of 23
Figure 1. Schematic draws of the Ulvac ERA-1000 CVD cluster tool. Reactors 1
and 2 are pumped by Roots pumps RBP1 and RBP2 backed by
mechanical pumps RP1 and RP2 during processing, while they are
pumped to base pressures prior to process by turbopumps TMP1 and
TMP2, backed by mechanical pump RP4. Mechanical pump RP3 is used
to evacuate the load lock. The various chambers are isolated from each
other by slit (gate) valves V1, V2, and V3.
TMP2
RP4
TMP1
CENTRALWAFERHANDLER
REACTOR 2 REACTOR 1
LOADLOCK
TOSCRUBBER
V1V2
V3
PROCESSPUMPS
RBP1 &
RP1
PROCESSPUMPS
RBP2 &
RP2
RP3
V11
V41
ulvacsim.paper.doc 08/20/99 p. 18 of 23
Figure 2. Variation of pumping speed with system pressure for turbo and
mechanical pumps.∗
∗ A representative plot of the pumping speed as illustrated in ALCATEL ATP Catalog
Pu
mp
ing
sp
eed
100
Mechanical PumpTurbo Pump
10210-210-410-610-810-12 10-10
System pressure ( Torr )
ulvacsim.paper.doc 08/20/99 p. 19 of 23
Figure 3. Comparison of reactor pressure and valve state signals during the initial
part of the process cycle, i.e., wafer loading into the reactor for
deposition. Data points are from dynamic simulation, while solid lines
are experimental measurements using equipment state signal
integration under LabView. Voltage signals for valve states (∼3.0V) are
scaled up by 100X for clarity. Valve state data from the simulation is
scaled vertically to match amplitude of experimental signals.
60 80 100 120 140
0
100
200
300
400
500V11 closesfor wafer load
deposition starts
valve opens forreactor pumpdownbefore deposition rise in reactor press
during wafer transferfrom CWH to reactor
reactor pressure valve V11 status valve V41 status
Pre
ssu
re (
mT
orr
)V
alve
sta
te in
(vo
ltag
e si
gn
al*1
00)
V
Process time (s)
ulvacsim.paper.doc 08/20/99 p. 20 of 23
Figure 4. Comparison of simulated deposition rate with experimental results
reported by Rosler et. al. 16.
1.2 1.4 1.6 1.8100
1000
10000 experimental data
simulation results
Dep
osi
tio
n r
ate
in A
/min
1000/T (K-1)
ulvacsim.paper.doc 08/20/99 p. 21 of 23
Figure 5 (a). Analysis of deposition rate and WF6 conversion vs. temperature for
different reactant ratios (H2:WF6 flow rate ratio = 20:1, 10:1, 5:1) for
0.2 Torr pressure.
350 400 450 500 550 6000
10
20
30
40
50
Process temperature (°C)
Pressure: 0.2 TorrH2 flow rate: 200 sccm
WF6 = 10 sccm (20:1 ratio)WF6 = 20 sccm (10:1 ratio)WF6 = 40 sccm (5:1 ratio)
Co
nve
rsio
n f
or
WF
6 (%
)
350 400 450 500 550 6000
1000
2000
3000
4000
5000
6000
7000
WF6 = 20 sccm (10:1 ratio)
Process temperature (°C)
WF6 = 10 sccm (20:1 ratio)
WF6 = 40 sccm (5:1 ratio)
Dep
osi
tio
n r
ate
(A/m
in) Pressure: 0.2 Torr
H2 flow rate: 200 sccm
ulvacsim.paper.doc 08/20/99 p. 22 of 23
Figure 5 (b). Analysis of deposition rate and WF6 conversion vs. temperature
for different reactant ratios (H2:WF6 flow rate ratio = 20:1, 10:1, 5:1)
for 0.5 Torr pressure.
350 400 450 500 550 6000
1000
2000
3000
4000
5000
6000
7000
WF6 = 20 sccm(10:1 ratio)
Process temperature (°C)
WF6 = 10 sccm(20:1 ratio)
WF6 = 40 sccm(5:1 ratio)
Dep
osi
tio
n r
ate
(A/m
in)
350 400 450 500 550 6000
10
20
30
40
50
WF6 = 20 sccm(10:1 ratio)
Process temperature (°C)
WF6 = 10 sccm(20:1 ratio)
WF6 = 40 sccm(5:1 ratio)
Co
nve
rsio
n f
or
WF
6 (%
)
Pressure: 0.5 TorrH2 flow rate: 200 sccm
Pressure: 0.5 TorrH2 flow rate: 200 sccm
ulvacsim.paper.doc 08/20/99 p. 23 of 23
Figure 6.Dynamic simulation (triangles) and experimental data (smooth curve) for
pressure burst observed in reactor during post-deposition wafer unload.
625 650 675 700 725
0
200
400
600
800
1000
wafer unload
pressure burst due toflushing of MFC valves
deposition processpressure = 900 mtorr
Rea
cto
r p
ress
ure
(m
To
rr)
Process time (s)